Enhancing Multi-Camera People Tracking with Conflict-aware Cosine Tracking and Human Matching Re-assignment

Multi-camera people tracking is a critical task in computer vision, particularly for applications such as surveillance, smart cities, and crowd monitoring. In this paper, we propose an enhanced model to address the problem of people tracking across large spatial areas with multiple cameras having overlapping fields of view. Our method combines the Conflict-aware Cosine Tracking (CC) algorithm with a Human Matching Re-assignment technique to improve identity association across different camera views. Compared to existing methods, our approach demonstrates better performance in tracking partially occluded objects and reduces the chances of misidentifying other objects as people. Key evaluation metrics, including ID Switch (IDS) and False Positive (FP), highlight the advantages of our method. Specifically, our model achieves over 85% performance in 4 out of 5 scenes in the ICGLab6 dataset, surpassing results from prior works [1] and [2] in most scenes. By integrating appearance features and 3D ground plane information, our approach enhances identity consistency across multiple camera views and significantly reduces False Positives. Experimental results demonstrate the effectiveness of our method in improving identity persistence and robustness in multi-camera people tracking systems, offering a promising direction for future advancements in this area.